00 China International Conference on Electricity Distribution Power ransission ower ype Deterining Method Based on Aerial Iage Analysis Jun Zhao,, Shu-tao Zhao and Qi-neng Jiang 3 School of Electric and electronic Engineering of North China Electric Power Uniersity, Baoding 07003, China Power Supply Bureau of Ordos 00406, China 3 Power Supply Bureau of Jiang Men, Jiang Men 59000, China E-MAIL: zj00@63.co, shutaozhao@63.co, and jqneng@6.co Abstract: Using helicopter aerial inspection of power transission line is an effectie easure in coplicated geographical enironent. At the present tie,the inspection of the aerial iage rely on anual obseration, its efficiency is low and undetected always happen by istake or careless. An autoatic analysis ethod by coputer has been proposed to process aerial ideo. After pretreatent, the iage quality has iproed. Edge detection is utilized to deterine the target area of the iage, and tower contours is obtained by cure fitting. Based on ariety tower features, teplate atching ethod is used to identify the power transission tower type. he analysis results of the aerial iage showed that the identification ethod is feasible. It can greatly reduce the labor burden of inspection, and iproe the efficiency of aerial iage analyzing. Keywords: Aerial iage; iage processing; cure fitting; feature atching. Introduction ransission line tower is the infrastructures in the power transission syste, and it is one kind of key equipents in electric energy transitting. o a large extent, the running of power transission syste is deterined by the towers working condition. ransission line corridor cross-oer unanned area, traffic and counication blind area, this lead to the difficulty of transission line periodically inspecting. An effectie easure has been proposed that using helicopter aerial inspection of power transission lines in coplicated geographical enironent. Aerial inspection iages contain a large nuber of transission lines running status inforation. At the present tie,the aerial iage analyzing ainly rely on anual obseration, its efficiency is low and undetected always happen by istake or careless.. he Process of ower Iage Analysis ower is supporting oerhead power transission wires and grounding wire line, and ake between wire, ground or across the property to aintain a certain distance. ower running state close to its shape changes, so soe tower-related faults can be deterined fro the aerial iage feature, and transission line auto inspection leel can be greatly increased. Howeer, the aerial iage background is ery coplex, there are any targets include towers, insulators, wire and identification cards, etc., it is ery difficult to achiee autoatic identification in fact. With the deelopent of coputer and iage processing technology, according to aerial iages tower color, grayscale, and structural characteristics, power tower fault autoated diagnosis based on aerial iage recognition is presented in this paper. Deterining the tower types is the first step of autoatic diagnosis the transission line faults. As shown in Figure, the transission line ideo hae been captured by digital caera, and the iage frae containing the independent tower can be obtained for the ideo. In the process of iage acquisition frequently leads to iage degradation. After the iage frae can be get fro aerial ideo strea, and then pre-process about the
00 China International Conference on Electricity Distribution iage background color and grayscale has been done, in order to reoe the specific color fro color iage. Cobined with the characteristics of the iage gray tower, the gray part irreleant with the tower is reoed. Aerial Iage Acquisition Power ower Deterining Power ower ypes Aerial Iage Pre-process ower Head Feature Matching Aerial Iage Edge Detection Linear Profile Cure Fitting ower Feature eplate Production Figure. he iage process of transission towers type deterining And then waelet transfor and cobine with edian filter is utilized to achiee de-noising caused by the process of iage capture. And then iage histogra equalization is utilized to be strengthened the iage edge inforation with the iage de-noised in pre-process. he next step is to detect aerial tower iage edge parts. he iage edge points are detected firstly, and the contour line can be deterined fro edge point s distribution. hrough coparing the corner edge point s nuber and detected straight edge nuber, soe useless inforation about recognized target. Paraeter of the tower edges fro the iage is calculated, this can be used to fit tower s contour cure. Each tower head has its own iage features, these features would be ade atching teplates. he extracted tower contour feature is copared with the teplate, the type of towers can be deterined. In the iage analysis, the tower feature exacted and atching teplate selection are extreely significant. 3. ower Iage Recognition Process 3. Iage Pretreatent In the aerial iages, the color and edges of the power tower ay be blurry, target is coered by ideo subtitles, een tower profile loss which cause difficulties to the equipent type recognition. So the iage preprocess ust be done to iproe the iage quality, soothing filter is used for blurring noise reduction in the iage analysis. Each pixel becoes a weighted aerage of its neighbors. he aerial iages of three colors including red, green and blue, calculates the histogra, or pixel distribution, take the appropriate color threshold color iages canbe transfored to grayscale. he prior knowledge is useful in the color transforing. According to the histogra of the iage, the entropy of the histogra of an iage with gray leels in the range [0,N-] is gien by H h( i) log h( i) h( i) h( i) in in N e N i0 i0 i0 i0 where i is the gray leel alue, h(i) represents the nuber of pixels in the iage at each gray leel alue, N is the total nuber of gray leels in the iage. It signifies the aount of with the black and white pixels inforation associated with the histogra. he optial threshold (H) is gray leel alue that axiizes the entropy, as () shows. 0 f ( x, j) H 50 gx 55 f ( x, j) H 50 ( ) () ower grey leel is between 00 and 50 in the grey iages, after procees as (), the backgroud interference is weakened (i.e., black for foreground and white for background). After this course, the calculation interal can be reduced greatly. 3. ROI Deterined by Cure Fitting Use fitting ethod contour to transfor the aerial tower linear cure into a contour cure and it is can be utilized one of the teplate atching features. In the process of cure fitting, use least squares ethod to fit the existing points of a straight line. he ethod of fitting least squares is to rely on the known data. First, the paraeters f x a a a ;,, () should be constructed to reflect the function of containing the unknown and n data points x, y, i,, n deiate fro the leel as forula (3). n,, ;,, J a a a y f x a a a (3) i i then apply the atheatical ethods to deand iniu of function J a a a cin J a, a, a,,, here the alue of a, a, a is pending alue what it isdeand. Such a set of alues akes the function i i
00 China International Conference on Electricity Distribution 3 ;,, f x a a a and n data points the closest in the A a a a (,, ) and Y y y y n (,, ) second square and the sense. Figure is the fitting results of cathead-tower head plans first characteristic cure. A. ower straight contour B. Cure fitting of tower Figure. owers straight line contour fitting results In the Figure A, Straight line fitting process is as follows: he straight line equation of tower contour is obtained preiosly, the for of using linear least squares cure fitting function containing unknown paraeters like: f x; a, a, a a r ( x) a r ( x) a r ( x) (4) When r( x), r( x),, r ( x) is linearly independent, can be reersible, the equations R RA RR R Y hae a unique solution. According to the linear theory of least-squares fitting paraeters, the solution of atrix A constituting deanded paraeters can be gained, using the coand in MALAB A \ paraeters directly. R Y 3.3 eplate Matching, obtain the deand In the recognition process of pole and tower Iage types, it should separate the iage features of each tower, using these characteristics to distinguish it with the other towers. he difference between cathead-towers and glass towers and other towers lies in the head shape, so in the recognition process, it should be atched in accordance with its characteristics of tower head. As for the corner tower, its characteristic is the corner tower cross ar truss structure. According to these features, the teplate of iage atching has been producted. where, rk ( x ) is a set of functions to be fitted and linear equations, k,,, n, n. a is deterined coefficient, k o seek a, a, a, ake the function reflecting the deiation J a a a,, the iniu, only need to use the necessary conditions J 0, k,, then a get the linear equations about a, a, a : k n r ( xi ) akrk ( xi ) yi 0 i k n r ( xi ) akrk ( xi ) yi 0 i k A.Ports of cathead-tower edge feature teplate he equations can be siplified to including: r ( x ) r ( x ) r ( x ) r ( x) r ( x) r ( x) R r( xn ) r( xn ) r ( xn ) R RA R Y,
4 00 China International Conference on Electricity Distribution included angle Y, so it is only when u, cy, then B. Ports of glass-tower edge feature teplate C. Ports of corner-tower edge feature teplate Figure 3. Soe tower edge feature teplate In contrast to the siilarity of cure and fitting cure in teplates, the algorith of NPROD is used to easure siilarity: N N i j i, j i, j R ( (5) N N N N i j Yi, j i j i j With the inner product for, it can be expressed: Y R( or ( )( Y Y ) Y Y R( (6) Y It can be seen fro the aboe equation, noralized product is actually the cos between ector u, and hae R(, here c is a constant. Otherwise R(, therefore, NPROD algorith is not subject to scale factor errors. Deterine the siilarity of 80% or ore cures that atching of the characteristics is successful, the tower type is identifyed. ake the iage of the sae location at different ties to easure the line profile paraeters and copare the results, and then judge the state of the tower. 3.4 he Result Of Matching In order to testify its feasibility, type classifications of cathead-tower iages, 4 Glass towers iages and 3 corner tower iages. hese Iages will be carried out. cathead-tower will be nubered as M-M. Glass tower will be nubered as J-J4. (J includes a glass tower and two cathead-tower, J3 consists of two glass towers). Corner tower will be nubered as Z, Z and Z3. According to the tower features, the tower is nubered according to its features. Such as: cathead-tower is characterized by a-a5, glass-tower is characterized by b-b5, corner-tower is characterized by c and c. hey hae 80% siilarity. ABLEⅠshows part of the tower iage and the teplate feature atching results. ABLEⅠ ower feature atching data analysis sheet ower ype Cathead-tower(M-M) Glass-tower(J-J4) Corner-tower(Z-Z3) Matching characteristics of the teplate nuber he nuber of correct atches he correct atch rate he nuber of correct atches he correct atch rate he nuber of correct atches he correct atch rate a 0 90.9% 0 0 0 0 a 00% 50% 0 0 a3 8 8.8% 50% 0 0 a4 9 86.4% 50% 0 0 a5 0 90.9% 50% 0 0 b 0 0 5 00% 0 0 b 0 0 5 00% 0 0 b3 0 0 5 00% 0 0 b4 0 0 5 00% 0 0 b5 0 0 5 00% 0 0
00 China International Conference on Electricity Distribution 5 c 0 0 0 0 3 00% c 0 0 0 0 3 00% Wine glass tower and the corner tower feature atching results for all atches correctly. It is due to the lack of tower saple. Howeer, cathead-tower feature atches correctly ore than 80%. In the actual recognition process, as long as two or ore features hae been identified, the type of tower can be identified. Based on this principle, in addition to J, iages of the reaining 8 towers can be recognized correctly. As Figure J contains two cathead-tower and a glass tower, tower in larger quantities and the resolution is not clearly. While J3 (includes two glass towers) atching results show that the iage contains ore than one tower case, this identification ethod also be applied. 4. Conclussions Achiee the power transission lines to autoatic inspection aerial iages can greatly reduce the burden on the labors. his paper presents a new ethod to identify the iages of the transission equipents, it preprocess the aerial iages with the coputer, cobination the characterize features of the towers in the iages to detect and use teplates atching ethod for the aerial iages recognition, to deterine the type towers. hrough the actual aerial iages analysis deonstrated the feasibility of this recognition algoriths, and further study the feasibility of an autoatic inspection aerial iages, transission lines fault detection and rapid response serice network with unexpected incidents. References H.B. Yan, Shanxi Power Grid Analysis of helicopter transission line inspection operations, Friends of Science and echnology, Vol 4, No. 8, pp.4-5, Aug. 008. Z. K. Sun, Z. K. Shen, Digital iage processing and application, National Defense Industry Press. 985. Brown. L.G, A surey of iage registration techniques, ACM Coputer Sureys, Vol 6, No.6, pp. 35-376, Jun. 99. G.J Yan, C.Y. Li, Autoatic Extraction of Power Lines Fro Aerial Iages, IEEE Geo-science and Reote Sensing Letters, Vol. 4, No. 3,pp. 387 39, July. 007. A. Mc. Andrew, Introduction to Digital Iage Processing with MALAB, hoson Learning, Inc, 004. D. Y. ong, Z. Y. Li, H.L. Zhang, Least-squares cure-fitting and MALAB ipleentation, Science and echnology Foru, Heilongjiang Science and echnology Inforation, 009. W. Quan.. D. Wang, Iage Matching Siilarity Measure Coparison and Analysis. Vol 38, No. 5, pp. 8-0, May 008.